JP3473750B2 - High-speed motion search device - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-speed motion search apparatus, and more particularly, to a high-speed motion calculating apparatus for compressing and coding a moving picture at high speed to obtain an optimum motion vector to improve compression efficiency. The present invention relates to a vector search device.

[0002]

2. Description of the Related Art Motion-compensated interframe coding is an operation for predicting a pixel Xijk of a k-frame to find a pixel value Xlm closest to Xij in a signal of a k-1 frame. In particular, this means an operation of searching for a block having the highest correlation in block units divided into fixed rectangular areas. Here, the spatial deviation between Xij and Xlm | i
-L | and | j-m | are transmitted as additional information (motion vector).

Conventionally, in an image compression apparatus for performing inter-frame predictive coding, a method of obtaining the difference with the most highly correlated portion and reducing the amount of data to be encoded is an effective means for improving the image quality. can give.

According to this method, the wider the vector search range, the more accurate the search can be performed, but the amount of processing increases dramatically. When high-speed compression processing is required, it is common to reduce the calculation amount by narrowing the search range or stopping the search when a part with a high degree of correlation is found. , First
Japanese Patent Application Laid-Open No. 10-271514, which is cited as a conventional example of the above.

However, when these methods are executed, there is a high possibility that the search process will be interrupted despite the existence of a portion having a higher degree of correlation, resulting in a problem that the image quality deteriorates. is there.

Further, Japanese Patent Laid-Open No. 8-32969, which is a second conventional example, describes a method of dynamically changing the search range by using the correlation with the same position of the previous frame.

In the method according to the second conventional example, since the search range does not reflect whether or not the image has a large motion, there is a possibility that a wide range is searched even in a small motion area, and the processing time is shortened. Not necessarily. For example, in the search of a part where the boundary part of the object has moved, the correlation at the same position as the previous frame is low, so a wide range is searched.

Further, Japanese Patent Application Laid-Open No. 10-191352, which is a third conventional example, describes a method of performing the processing for obtaining the correlation in parallel, but since it is necessary to have a plurality of arithmetic units. There has been a problem that the circuit scale becomes large and a plurality of CPUs are required when it is realized on a personal computer, for example.

[0009]

However, the above-mentioned first to third prior arts have the following drawbacks, respectively.

That is, the first conventional example has a drawback that the image quality is deteriorated because the possibility that the search process is interrupted increases even if there is a portion having a higher correlation.

Further, in the second conventional example, since the search range does not reflect whether or not the image has a large motion, there is a high possibility that a wide range is searched even for a small motion part, and the processing time is shortened. Not necessarily. For example, in the search for a part where the boundary of the object has moved, there is a drawback that a wide range is searched because the correlation at the same position as the previous frame is low.

The third conventional example has a drawback that the circuit scale becomes large because it has to have a plurality of arithmetic units, and for example, a plurality of CPUs are required when it is realized on a personal computer.

In view of the above-mentioned conventional circumstances, the present invention has been made to solve the above-mentioned drawbacks inherent in the prior art. Therefore, the object of the present invention is to detect a motion vector accurately at high speed. It is to provide a novel motion vector search device that enables the above.

[0014]

In order to achieve the above object, a high speed motion retrieval apparatus according to the present invention comprises an image input unit for inputting an image in units of one frame and a current frame output from the image input unit. And a previous frame and a motion search means for obtaining a correlation between the motion vector and a previous frame, and a difference calculation with a block at the position of the reference frame using a vector to a position where a minimum prediction error is obtained by the search by the motion search means as a motion vector. Inter-block difference means for performing, frequency conversion means for converting the block difference data obtained by the inter-block difference means into frequency components, quantization means for quantizing the frequency components, and quantization output of the quantization means. In a compression coding apparatus having a variable-length coding means for compressing and outputting the motion vector, the first-stage motion searching means and the second-stage motion are used as the motion vector searching means. Search means is provided, and the first-stage motion search means searches a search range determined by referring to the search result in the previous frame, and searches this range if the minimum prediction error is smaller than the threshold value. If the prediction error is larger than the threshold value, the motion search process is terminated at the point of time, and the part not searched by the first-step motion search means within the search range preset by the second-step motion search means. To search for a location where the prediction error is minimized, perform the search in two stages, and dynamically change the search range in the first stage to perform a high-speed motion vector search with few false detections. Is characterized by.

When the image data is input to the image input means, it is determined whether or not the search information data can be referred to. However, since the information cannot be referred to in the first search, the first stage motion search means has a predetermined value. The search range is searched without any special processing.

As the threshold value, a search result of macroblocks located at the same position in the previous frame is used.

The search range of the first-stage motion search means is determined in a square area having a side that is twice the length of the motion vector obtained in the previous frame.

When the search information can be referred to, the motion vector and prediction error information of the macroblock at the same position in the previous frame is obtained, and the search start position for that vector is calculated based on the motion vector information of the previous frame. The location where the prediction error is minimized is predicted by shifting the search, and the search is performed centering on that location.
The range to be searched for the movement of the stage is determined.

The search range may be either the maximum value or median value of the vectors of the surrounding blocks already searched in the frame, or the maximum value or median value of the vectors at the same position in the past several frames. If the minimum value of the prediction error obtained in the search range is smaller than the prediction error in the same-position block of the previous frame, the search process ends, otherwise the search process continues.

Search information storage means is provided after the output of the first-stage and second-stage motion search means, and is obtained by the first-stage or second-stage motion search means for motion search of the next frame. The motion vector value and the minimum value of the prediction error are stored in the search information storage means.

The first stage motion search means and the quantization means dequantize the output of the quantization means and further perform inverse frequency conversion to convert frequency components into pixel components, and refer to the pixel components. Motion compensation means is provided for generating frame data.

[0022]

In FIG. 1, the first stage motion search means 12 searches the search range determined by referring to the search result in the previous frame, and if the minimum prediction error is smaller than the threshold value, this range is searched. The motion search process ends when the search is performed.
If the prediction error is larger than the threshold value, the second-step motion search means 13 searches for a portion within the preset search range that has not been searched for in the first-step motion search, and the prediction error is minimized. Ask for a place.

By thus performing the search in two stages and changing the search range in the first stage, it is possible to perform a high-speed motion vector search with less erroneous detection.

[0024]

BEST MODE FOR CARRYING OUT THE INVENTION Next, a preferred embodiment of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a block diagram showing an embodiment of the present invention.

[Structure of Embodiment] Referring to FIG.
One embodiment of a high-speed motion search device according to the present invention is an image input unit 11, a first-stage motion search unit 12, a second-stage motion search unit 13, a search information storage unit 14, and an inter-block difference unit. 15, frequency conversion means 16, quantization means 17, dequantization means 18, and inverse frequency conversion means 19
, Motion compensation means 20, variable length coding means 21, and output means 22.

Each of these means operates roughly as follows.

An image is taken in from the camera or the like by the image input means 11 in units of frames, and the first stage motion retrieval means 12
Then, the correlation between the previous frame and the current frame stored as the reference frame in the motion compensation means 20 is obtained. The search range at this time is a square area having one side that is twice the length of the motion vector obtained in the previous frame. If the minimum prediction error obtained by searching this search range is smaller than the prediction error of the search result of the macro block (coding block unit) at the same position in the previous frame, the second stage motion search means 13 The inter-block difference means 1 performs the difference calculation with the block at that position of the reference frame without using the search by
Done at 5.

Otherwise, that is, when the minimum prediction error obtained by the first stage motion search means 12 is larger than the prediction error of the search result of the macroblock at the same position in the previous frame, The second stage motion search means 13 searches for a portion within the prescribed search range that is not searched by the first stage motion search means 12, and the vector to the place where the minimum prediction error is obtained is set as the motion vector, The difference calculation between the reference frame and the block at that location is performed by the inter-block difference means 15.
Done in.

Further, the motion vector value and the minimum value of the prediction error which have been obtained for the motion search of the next frame are stored in the search information storage means 14.

The obtained block difference data is converted into frequency components by the frequency conversion means 16, and the quantization means 17 is used.
Quantization is performed in. The quantized data is encoded into a compression code by the variable length coding means 21, and output means 22.
Is output with.

In order to create a reference frame for the next frame compression, the result of the quantizing means 17 is the inverse quantizing means 1.
Inverse quantization is performed in 8, inverse frequency conversion means 19 converts the frequency components into pixel components, and motion compensation means 20 generates reference frame data.

[Operation of Embodiment] Next, the operation of the embodiment will be described in detail with reference to the drawings.

FIG. 2 is a flow chart showing an example of the operation flow of this embodiment.

Referring to FIG. 2, when the image data is input, it is confirmed whether the search information data can be referred to (step S11).

Since this information cannot be referred to in the first search, the default search range is searched without any special processing (step S12).

When the search information can be referred to, the information of the motion vector and prediction error of the macroblock at the same position in the previous frame is acquired (step S13).

Since the direction of motion rarely changes abruptly, the position where the prediction error is minimized is predicted by shifting the search start position by that vector based on the motion vector information of the previous frame. , The search is performed mainly on that place (step S14).

The range to be searched is determined from the acquired vector information, and twice the vector length is set as the search range.
The first search is performed (step S15).

Here, the search range is assumed to use vector information at the same position in the previous frame, but this may be the maximum value or median value of the vectors of the already searched surrounding blocks in the same frame, or past several frames. It may be the maximum value or median value of the vector at the same position of. If the minimum value of the prediction error obtained in this search range is smaller than the prediction error in the same-position block of the previous frame, the search process ends, otherwise the search process continues (step S16).

When it is determined that the search is continued, the second-stage search processing is performed (step S17).

In the search here, an unsearched area is searched for in the first search process within the predetermined search range. If a result smaller than the prediction error obtained in the first search is obtained, the vector value is updated. Otherwise, the first search result is determined by using the vector value of the current macroblock.

Vector values obtained by these processes,
Since the prediction error value is referred to when the next frame is searched, it is stored in the temporary storage area (step S18).

Next, the present embodiment will be described using a specific example.

In particular, here, MPEG (Moving)
The Picture Experts Group) will be described.

The frame structure of MPEG1 will be described with reference to FIG.

Referring to FIG. 3, frame types include
There are three types: I-frames that are not predicted, P-frames that are predicted from past information, and B-frames that are predicted from past / future information.

In this example, the minimum number of frames required for compression is four frames from the reference frame to the next reference frame, which is IBBP or PBBP.

Considering the reference direction and order, ref1
Indicates that there is no reference result for the previous frame, search information cannot be acquired, and a predetermined range is searched, and ref2 determines the search range and threshold value of the first stage based on the search information of ref1 and performs a search. ref3 searches based on the search information of ref2.

Next, when a search for ref4 is performed, B2
Since there is no information that predicts P0 to P0, a predetermined range is searched, and ref5 is searched based on the search information of ref4.

The ref6 can be referred to by using the prediction error information of ref4 as it is, the vector information as it is, and the information of the opposite direction. ref9
Also, since the reference information cannot be obtained for the same reason as ref4, a predetermined search is performed.

FIG. 4 is a diagram for explaining a motion search method.

Next, referring to FIG. 4, a method of motion search will be described. The motion vector search is performed by a method of obtaining a macroblock difference between a reference frame and a 16 × 16 rectangular area, and using a sum of absolute values or a sum of squares of the difference values as a prediction error to obtain a block with the smallest prediction error. It is common. Default search range is -15 vertically and horizontally
Set to +15.

In the first search, the vector of the macroblock at the same position in the previous frame is referred to and the search start point is moved by that vector. This is because the vector at the same position between adjacent frames is predicted not to change abruptly, and the starting point is moved to the center of the prediction to reduce the probability of false detection of the vector even if the search range is narrowed. Because there is a tendency.

If the current position is (32,32) and the reference vector (3, -4), the search start position is (3
5, 28). Since the search range is also determined from the reference vector length, the search range in the first stage is -5 to +5.

Here, the vector length is simply used as it is as the search range, but a 10% increment of the vector length may be used as the search range. As a result, the area surrounded by the dotted line in FIG. 4 is the search area in the first row, the area surrounded by the solid line is the default search area, and the area between the dotted and solid lines is the search area in the second row. .

If the search in the first step goes out of the default search range as shown in FIG. 5, the default area is given priority, and the protruding portion is excluded from the search target.

If the reference prediction error is 80 and the prediction error is as shown in FIG. 6 in the first search, the minimum prediction error value 75 is equal to or less than the reference prediction error value 80, and therefore the search process is performed in two steps. The process ends without performing any eyes, and the prediction error of this macroblock is stored as 75 for reference to the next frame.

Further, as shown in FIG. 7, when the minimum prediction error value of the first step is 82, it is larger than the reference prediction error value, so the second step is searched and both the first step and the second step are searched. The vector whose prediction error value is the smallest is defined as the motion vector.

[0060]

The present invention is constructed and operates as described above, and according to the present invention, the following effects can be obtained.

The first effect is that high-speed motion vector search can be performed.

The reason is that the number of times of block matching processing can be reduced because the search range is adaptively changed within a range that is always less than or equal to the specified value.

The second effect is that in the motion vector search process, the vector search can be performed at higher speed in a place with less motion.

The reason is that since a new search range is set by referring to the vector detected at the same position in the previous frame, the search range becomes narrower at the place where the motion of the previous frame is smaller, that is, the place where the vector is smaller. Is.

The third effect is that an accurate motion vector can be detected in the motion vector search process.

The reason is that macroblocks having high correlation can be accurately detected without aborting the search process until a certain range is searched.

The fourth effect is that in the above motion vector search processing, an accurate motion vector can be detected even if the magnitude of motion changes.

The reason is that an accurate motion vector can be detected by searching a predetermined search range if the correlation is low even if a certain search range is searched.

[Brief description of drawings]

FIG. 1 is a block diagram showing an embodiment according to the present invention.

FIG. 2 is a flowchart showing an operation example of an embodiment according to the present invention.

FIG. 3 is a diagram illustrating a frame structure of MPEG1 used in the present invention.

FIG. 4 is a diagram for explaining a search method according to the present invention.

FIG. 5 is a diagram for explaining a search method according to the present invention.

FIG. 6 is a diagram for explaining a search method according to the present invention.

FIG. 7 is a diagram for explaining a search method according to the present invention.

[Explanation of symbols]

11 ... Image input means 12 ... 1st stage motion search means 13 ... Second stage motion search means 14 ... Search information storage means 15 ... Block difference means 16 ... Frequency conversion means 17 ... Quantization means 18 ... Inverse quantization means 19 ... Inverse frequency conversion means 20 ... Motion compensation means 21 ... Variable length coding means 22 ... Output means

Claims (5)

(57) [Claims] 1. An image input unit for inputting an image in units of one frame, a motion search unit for obtaining a correlation between a current frame and a previous frame output from the image input unit, and a search by the motion search unit. An inter-block difference means for performing a difference calculation with a block at that location of a reference frame using a vector to a location where the minimum prediction error is obtained as a motion vector, and block difference data obtained by the inter-block difference means as a frequency component. In the compression encoding device, the compression encoding device includes: a frequency conversion unit that converts the frequency component into a frequency component, a quantization unit that quantizes the frequency component, and a variable length encoding unit that compression-encodes and outputs the quantized output of the quantization unit, As the motion vector search means, a first stage motion search means and a second stage motion search means are provided, and the first stage motion search means obtains the search result in the previous frame. If the minimum prediction error is smaller than the threshold value, the motion search process is terminated at the time when the minimum prediction error is smaller than the threshold value. The step which is not searched by the first step motion searching means within the search range preset by the step moving search means is searched to find the place where the prediction error is minimum, and the search is performed in two steps. , a fast motion search apparatus for searching for erroneous less motion vector of detected fast by dynamically changing the search range of the first stage, the search range of the first stage motion search means, the previous frame Request
A square with one side that is twice the length of the motion vector
A high-speed motion search device characterized by being a region . 2. When the search information can be referred to, information on a motion vector and a prediction error of a macroblock at the same position in the previous frame is acquired, and the vector search is performed based on the motion vector information of the previous frame. By predicting the place where the prediction error is the minimum by shifting the start position, the search is performed centering on that place, and 1 is obtained from the acquired vector information.
The high-speed motion search device according to claim 1, further comprising: determining a range of motion search of the stage. 3. The search range is either the maximum value or median value of vectors of surrounding blocks already searched in a frame, or the maximum value or median value of co-located vectors in the past several frames. , If the minimum value of the prediction error obtained in the search range is smaller than the prediction error in the same position block of the previous frame, the search processing ends,
The high-speed motion search device according to claim 2 , further comprising continuing the search process otherwise. 4. A search information storage means is provided between the input of the first-stage motion search means and the outputs of the first-stage and second-stage motion search means, and the first-stage is used for motion search of the next frame. The high-speed motion search device according to claim 1, further comprising storing the motion vector value and the minimum value of the prediction error obtained by the first and second stage motion search means in the search information storage means. 5. Between the first-stage motion search means and the quantization means, the output of the quantization means is inversely quantized, and further the inverse frequency conversion is performed to convert the frequency component into the pixel component, The high-speed motion search device according to claim 1, further comprising motion compensation means for generating reference frame data from the pixel component.